CN107197006B - Multi-constraint service selection method and device based on global QoS decomposition - Google Patents

Abstract

The invention relates to a multi-constraint service selection method based on global QoS decomposition and a device thereof, which is realized by establishing a single-target optimization model with multiple constraint conditions, and the method comprises the following steps: establishing a corresponding dependency set and a conflict set for each candidate service according to the service dependency relationship transfer characteristics; decomposing the global QoS constraints into local QoS constraints corresponding to each service class; filtering candidate services which do not meet the local QoS constraint under the service class; checking all the filtered candidate services and updating the dependency set and conflict set of the remaining candidate services; performing combined replacement of the quality scales in a non-solution state by a self-adaptive replacement method; calculating the local fitness of the candidate service; and selecting the candidate service with the maximum local fitness in each service class to form the final combined service. The invention is greatly optimized in complexity and running time, meets the real-time requirement of users, reduces the scale of candidate service space, and effectively ensures the quality and performance of network combination service.

Description

Multi-constraint service selection method and device based on global QoS decomposition

Technical Field

The invention belongs to the technical field of computer networks, and particularly relates to a multi-constraint service selection method and device based on global QoS decomposition.

Background

As a novel distributed Computing mode, a Service-Oriented Computing (SOC) technology can seamlessly and dynamically combine various Web services distributed in a heterogeneous network to form a large-granularity combined Web Service so as to meet increasingly complex requirements of users. With the prosperity and development of the Web market, a large number of Web services on the network have overlapped functional attributes, so that the user no longer meets the requirements on functions when selecting services, and also puts forward requirements on the quality of service (QoS for short). However, due to various uncertainty factors of the network environment, some services may have a failure in function or a QoS mutation during the operation process, and at this time, the method capable of quickly selecting the service is significant for ensuring the quality and performance of the composite service. At present, a great deal of research is focused on service selection under the global QoS constraint condition, and the existing QoS-based service selection method has the following disadvantages: (1) in general, in the user requirement for service composition, the user only gives the end-to-end QoS constraint of the composition service, and does not give the local constraint for each basic service class. At present, a large number of schemes are based on global QoS constraint, and an intelligent evolution algorithm is used for optimally searching combined services, but the algorithms are generally high in computational complexity, and the running time of the algorithms is seriously dependent on the scale of a candidate service set. Once the number of Web services in a network increases, the runtime doubles and it becomes difficult to meet the real-time requirements of users. (2) The existing service selection methods only consider the global QoS constraint proposed by the user, but ignore the functional constraint relation possibly existing between the candidate services. In a massive Web service environment, different candidate services can be interdependent or conflict with each other due to problems of business association, technical compatibility and the like. A hill climbing repair operator is introduced into a genetic algorithm in part of schemes, the hill climbing repair operator can repair the schemes which do not meet conflict dependency constraints among services in a population, so that the optimization direction of the algorithm is guided, but when the constraint scale among the services is increased, the repair time is multiplied, and the real-time requirements of users are difficult to meet. Part of the schemes provide a multi-constraint service selection method based on local approximate filtering, which quickly filters candidate service spaces through global QoS constraint and functional constraint among services, and then searches out the best combined service in the rest candidate services by utilizing a directed particle swarm algorithm.

Disclosure of Invention

Aiming at the defects in the prior art, the invention provides a multi-constraint service selection method and a device thereof based on global QoS decomposition.

According to the design scheme provided by the invention, the multi-constraint service selection method based on global QoS decomposition is realized by establishing a single-target optimization model with multiple constraint conditions, and the realization process comprises the following steps:

step1, establishing a corresponding dependency set and a conflict set for each candidate service according to the transfer characteristics of the service dependency relationship;

step2, decomposing the global QoS constraint provided by the user into local QoS constraints corresponding to each service class through a CGA culture genetic algorithm; filtering candidate services which do not meet the local QoS constraint under the service class;

step3, checking all the filtered candidate services, if the candidate services are in the dependence set of the filtered candidate services, removing the candidate services from the dependence set, and updating the dependence set and the conflict set of the remaining candidate services;

step 4, performing non-solution quality scale combination replacement on candidate services in the service class by a self-adaptive replacement method, and dynamically adjusting a QoS constraint boundary;

step 5, calculating the service compatibility and local fitness of the candidate service based on the global QoS constraint and the functional constraint among the services;

and 6, selecting the candidate service with the maximum local fitness in each service class according to the local fitness calculation result of the candidate service in the step 5, and forming a final combined service according to the selected candidate service.

In the above, the single-target optimization model is represented as:

wherein w is_k(k is more than or equal to 1 and less than or equal to r) is the weight of the QoS attribute k, and satisfies

CS is a composite service composed of m basic service classes Si (1 ≦ i ≦ m), and is denoted as CS { S1, S2, … …, Sm }; x is the number of_ijRepresentative service class S_jThe selection state of the ith candidate service; global QoS constraint relation C ═ C₁,C₂,......C_rDenotes C_kRepresenting global constraints of the kth QoS Attribute, q_k(CS)≤C_kIndicating that the k-th QoS attribute aggregate value of the composite service is required to meet the corresponding global constraint; the functional constraints between services are respectively defined by a dependency set T ═ T₁,t₂,......t_kD ═ D } and the set of conflicts₁,d₂,......d_uDenotes, dependency relationships

Representing service class S_aThe function of the b-th candidate service in (a) depends on the service class S_cThe d-th candidate service in (a); conflicting relationships

Representing service class S_eFunction and service class S of the f-th candidate service in (1)_gThe functions of the h-th candidate service in (a) conflict with each other.

As mentioned above, the step1 includes the following steps: the dependency delivery rules are as follows:

and s_d∈d(s_c) (3)

And s_b∈d(s_c) (5)，

Formula (2) represents s_bDependent on s_aThen s_bIs incorporated into s_aDependent set of (t)(s)_a) Performing the following steps; formula (3) represents s_cAnd s_dIf the two conflict with each other, the two conflict with each other and are included in the conflict set of the other party; formula (4) represents s_bDependent on s_a，s_aDependent on s_cThen s_bIs incorporated into s_cDependent set of (t)(s)_c) Performing the following steps; formula (5) represents s_bDependent on s_a，s_aAnd s_cConflict, then s_bAnd s_cThere is also a conflict.

In the above, the step2 is to decompose the CGA cultural genetic algorithm into the local QoS constraint corresponding to each service class, express the quality scale combination by constructing the chromosome model, each service class corresponds to one quality scale combination, embed the evolution operation of the CA genetic method into the population space of the CA cultural method, and introduce the collaborative learning mechanism into the belief space, and the global QoS constraint is decomposed into the local QoS constraint as follows:

step 21, randomly generating an initial effective solution in the population space according to the size of the population space, and evaluating all solutions through a fitness function;

step 22, performing evolution operation on the solution in the population space, wherein the evolution operation at least comprises selection operation, cross operation and variation operation;

step 23, selecting a better solution from the population space, transmitting the better solution to the belief space, and replacing the worse solution in the belief space according to the cumulative update times of the solutions in the belief space; performing collaborative learning operation on the solutions in the belief space, evaluating the newly generated solutions, reselecting a better solution from the belief space, and filtering out the rest solutions;

step 24, judging whether the current iteration times meet set conditions, if so, outputting an optimal solution in a belief space; otherwise, the loop iteration is performed by returning to step 22.

Preferably, the fitness function is expressed as:

，

wherein,

for the weight of the mass scale, the calculation formula is as follows:

for service class S_jZhongbelong to the quality staff gauge

Number of candidate services within, n (S)_j) Is service class S_jThe number of candidate services in (a) is,

is service class S_jZhongbelong to the quality staff gauge

Maximum utility function value of all candidate services within，Q_max(S_j) Is service class S_jMaximum utility function values of all candidate services; m is the total number of service classes, r is the total number of QoS attributes, and d is the total number of quality scales; the formula (11) ensures that the decomposed local QoS bundle can meet the global QoS constraint after aggregation; in the formula (12)Representing service class S_jMedium mass scale

Is selected, otherwise

It is guaranteed that under each QoS attribute of each service class, only one quality scale is selected.

As mentioned above, the step 23 performs the collaborative learning operation on the solution in the belief space, which includes the following steps:

step 1: randomly selecting t chromosomes from the belief space to form a cooperative learning group which is marked as

Wherein L is_i(1. ltoreq. i.ltoreq.t) represents a chromosome,

representing the jth gene of a chromosome as the service class S_jThe mass scale combination of (1);

step 2: comparing and learning the genes of each row of all chromosomes in the Group, and selecting the optimal genes of each rowI.e. from the service class S_jSelecting an optimal one of the t combinations of the quality scales;

step 3: recombining the optimal genes of each row to form a new chromosome, i.e.

Each service class S of the chromosome_jThe quality scale combinations of (a) are all within Group optimal.

Furthermore, in step 4, the quality scale combination replacement without solution state is performed by using an adaptive replacement method, and the QoS constraint boundary is dynamically adjusted, which includes the following contents: after filtering, if the service class has no candidate service, namely the situation of no solution exists, replacing the optimal quality scale combination of all the current service classes with the suboptimal quality scale combination in the belief space, and returning to the step2 for execution until no solution state appears.

As described above, in step 5: hypothesis candidate service s_jBelonging to the service class S_jThen, the calculation formula of the service compatibility is as follows:

，

wherein, t_i(s_j) Is a service s_jDependent set of (t)(s)_j) In the service class S_iThe number of services. d_i(s_j) Is a service s_jConflict set d(s)_j) In the service class S_iThe number of services. I S_iIs service class S_iThe total number of candidate services; the calculation formula of the local fitness is as follows:

u(s_j)＝con(s_j)×Q(s_j)，

，

con(s_j) Is s_jService compatibility of (C), Q(s)_j) Is the value of the utility function thereof,

is service class S_jThe maximum value on the k-th QoS attribute,

is service class S_jMinimum value on k-th QoS Attribute, q_k(s_j) Is a candidate service s_jValue on the kth QoS Attribute, w_kIs the weight of the kth QoS attribute given by the user.

A multi-constraint service selection device based on global QoS decomposition is realized by a single-target optimization model with multiple constraint conditions, and comprises: a function constraint establishing module, a local constraint decomposition module, a candidate service updating module, a non-solution state replacement module, a local fitness solving module and a final combined service selecting module,

the function constraint establishing module is used for establishing a corresponding dependency set and a corresponding conflict set for each candidate service according to the transmission characteristics of the service dependency relationship;

the local constraint decomposition module is used for decomposing the global QoS constraint provided by the user into local QoS constraints corresponding to each service class through a CGA culture genetic algorithm; filtering candidate services which do not meet the local QoS constraint under the service class;

the candidate service updating module is used for checking all the filtered candidate services, removing the candidate services from the dependency set if the candidate services are in the dependency set of the filtered candidate services, and updating the dependency set and the conflict set of the remaining candidate services;

the non-solution state replacement module is used for replacing the quality scale combination in the non-solution state by a self-adaptive replacement method aiming at the candidate service in the service class and dynamically adjusting the QoS constraint boundary;

the local fitness solving module is used for calculating the service compatibility and the local fitness of the candidate service according to the global QoS constraint and the functional constraint between the services;

and the final combined service selection module is used for selecting the candidate service with the maximum local fitness in each service class according to the local fitness calculation result of the candidate service in the local fitness solving module and forming the final combined service according to the selected candidate service.

In the above multi-constraint service selection apparatus, the local constraint decomposition module includes: a fitness function design unit, a chromosome model construction unit and a belief space learning unit,

the fitness function design unit is used for converting the global QoS optimal decomposition problem into a single-target optimization problem by designing a fitness function;

a chromosome model construction unit for representing quality scale combinations by constructing a chromosome model, each service class corresponding to one quality scale combination;

and the belief space learning unit is used for embedding the evolution operation of the CA genetic method into the population space of the CA culture method, introducing a collaborative learning mechanism into the belief space and acquiring the optimal quality scale combination of all the service classes.

The invention has the beneficial effects that:

the invention decomposes global QoS constraint into local QoS constraint, in order to achieve the best decomposition effect, the evolution operation of genetic algorithm is introduced into the population space of the genetic algorithm through CGA (Carrier-grade Advance) cultural genetic algorithm, a collaborative learning mechanism is introduced into a belief space, and the optimal quality scale combination of all service classes, namely the optimal local QoS constraint, is obtained through the iteration of double-layer space; secondly, filtering the candidate services which do not meet the multi-constraint condition by combining the dependency conflict relationship among the candidate services; then, performing state detection on the remaining candidate service spaces, if the situation that a certain service class does not have candidate services exists, executing a self-adaptive replacement strategy, replacing the current optimal quality scale combination with a suboptimal quality scale combination, and performing a double-layer filtering step again; and finally, calculating the local fitness of the residual candidate services, and obtaining the final combined service by a local optimal method. Compared with a global QoS service selection method, the scheme is greatly optimized in terms of computational complexity and running time, and can meet the real-time requirements of users; the method fully considers the functional dependency conflict relationship possibly existing between the candidate services, combines the functional constraint relationship with the QoS constraint relationship, filters the candidate services which do not meet the multi-constraint condition, reduces the scale of the candidate service space, greatly reduces the complexity of the algorithm, and effectively ensures the quality and the performance of the network combination service.

Description of the drawings:

FIG. 1 is a schematic flow diagram of the process of the present invention;

FIG. 2 is a schematic diagram of an implementation framework of a partial embodiment;

FIG. 3 is a block diagram of a CGA cultural genetic algorithm;

FIG. 4 is an exploded view of a mass scale;

FIG. 5 is a schematic diagram of a chromosome model;

FIG. 6 is a schematic diagram of a belief space learning process;

FIG. 7 is a schematic view of the apparatus of the present invention;

FIG. 8 is a diagram illustrating comparison of performance at different candidate service scales in a simulation example;

FIG. 9 is a graph showing comparison of performance at different constraint scales in a simulation example.

The specific implementation mode is as follows:

quality of Service (QoS): non-functional attributes representing Web services, including response time, reputation, availability, reliability, etc., are another important criteria for evaluating the quality of a service. Service compatibility: indicating the overall compatibility of a certain candidate service with other candidate class services. Local fitness: the degree of the candidate service in the service class is determined by the utility function value and the service compatibility. Genetic Algorithm (Genetic Algorithm, GA): the method is a global search algorithm derived by simulating evolution operations such as selection, intersection, variation and the like in nature by adopting the ideas of survival, excellence and disadvantage elimination of the Darwinian evolution theory. Culture Algorithm (Culture Algorithm, CA): a double-layer evolution model algorithm consists of a population space and a belief space, and simulates the evolution process of culture from a microscopic angle and a macroscopic angle respectively. Benefit-type QoS attributes: refers to the property that the larger the value is, the better the QoS is, such as reliability, availability, throughput, etc. Cost-type QoS attributes: refers to the attribute that the smaller the value is, the better the QoS is, such as response time, cost, etc. Dependence relationship: if the function of one candidate service depends on another candidate service, a dependency relationship exists between the candidate services. Conflict relationship: if the function of one candidate service conflicts with the function of another candidate service, a conflict relationship exists between the functions. And functional constraint: refers to a constraint formed between candidate services due to a functional dependency or conflict relationship. Global QoS constraints: refers to the constraints on the overall QoS for the composite service that are given by the user. Local QoS constraints: refers to QoS constraints for each basic service class that constitutes a composite service. Quality scale: each QoS attribute value range under the service class is divided into a plurality of discrete quality sets, and one quality set is called a quality scale. Adaptive replacement strategy: and when no candidate service exists in a certain service class after filtering, replacing the current optimal quality scale with a suboptimal quality scale, which is called as a self-adaptive replacement strategy.

The genetic algorithm mainly improves the adaptability of each individual through natural selection, intersection, variation and other genetic operations. The algorithm has a population consisting of a group of individuals, each individual is evaluated to be good or bad in the population evolution process and an adaptive value of the individual is obtained, and the individual evolves towards higher fitness under the actions of selection, intersection and mutation operators so as to achieve the aim of seeking the optimal solution of the problem. 1) Gene: also known as genetic elements, are molecular fragments that possess a large amount of genetic information. Genes are the basic genetic units for controlling the traits of organisms, which transmit their genetic information to their progeny through genes, which use the replication process to accomplish such transmission. 2) Chromosome: a substance consisting of a protein together with a DNA molecule of double helix structure and a small amount of RNA, all its genetic information of an organism is included in all chromosomes in each cell. 3) Population: each species is composed of a certain number of individuals, and the sum of all the individuals that make up the species is called the population. In the genetic algorithm, a population also contains a space of a solution of an actual problem in a certain generation and is also a set of possible solutions, and the population provides a genetic evolution search space for searching the solution for the genetic algorithm. 4) Fitness is as follows: the fitness is used for measuring the degree of goodness and badness of each individual in the population, and is also a standard for measuring the adaptability of each individual to the living environment of the individual. In the genetic algorithm, each individual chromosome in the population is coded firstly, and then the chromosome code of each individual is obtained, one individual is a possible solution of the actual problem, and all the possible solutions are in one-to-one correspondence with corresponding function values. 5) Selecting: in the genetic algorithm, the fitness is used as an index, and the individuals with high fitness in the current population are selected, so that preparation is made for the next genetic evolution operation. The greater the fitness, the greater the probability of selection and thus the greater the probability of inheritance to the next generation. 6) And (3) crossing: for two selected individuals (chromosomes) needing to be crossed, a cross point for cross exchange is given, then the two individuals are taken as parent individuals, and cross exchange is carried out at the cross point, so that two new offspring individuals (chromosomes) are generated after recombination, and the characters of the two new offspring individuals are combined by the characters of the parent individuals. 7) Mutation: the general process of mutation is to arbitrarily select an individual (chromosome) from a population, and then change a character at a certain position of the chromosome code of the individual with a certain probability, and whether the individual (chromosome) has mutation or not is controlled by the mutation probability. The method mainly comprises the following steps:

1) coding according to the parameter set of the problem to be solved;

2) initializing a group;

3) selecting individuals to be entered into a "generation" according to a rule determined by the fitness value of the individuals;

4) performing cross operation according to the cross probability;

5) carrying out mutation operation according to the mutation probability;

6) if some termination condition is not met, go to 3), otherwise go to 4);

7) and outputting the chromosome with the optimal fitness value in the population as a satisfactory solution or an optimal solution of the problem.

The cultural algorithm is a knowledge-based double-layer evolution system, which comprises two evolution spaces: one is a belief space composed of experience and knowledge acquired during evolution; and the other is a population space consisting of specific individuals, and self iterative solution is carried out through evolution operation and performance evaluation. 1) Population space: the process of individual evolution according to a certain behavior criterion is simulated from a microscopic perspective, and different computing architectures can be used for representing the population space. 2) Belief space: from the macroscopic perspective, the evolution processes of culture formation, transmission, comparison and the like are simulated, the belief space comprises environmental knowledge, standard knowledge, constraint knowledge and the like, and different knowledge types represent different characteristics of excellent individuals in the population. 3) An acceptance function: the population space transfers the individual experience to the belief space through the acceptance function, and actually provides a group of optimal subsets to the belief space, and the first few names are generally ranked according to a certain percentage in the optimization problem. 4) Influence function: after the belief space forms the group experience, the behavior rules of the individuals in the group space are modified through the influence function, so that the individual space obtains higher evolution efficiency. The population space and the belief space are two relatively independent evolutionary processes, and the two spaces are connected by a set of communication protocols consisting of an accept function accept () and an influence function infilue (). Individuals in the population space form individual experiences in the evolution process, and the individual experiences are transferred to the belief space through an accept () function. The belief space compares and optimizes the individual experience according to a certain behavior rule to form a population experience, and the belief space is updated by an update () function according to the existing population experience and the new individual experience; and the influence () function can guide the evolution of the population space by using the empirical knowledge of the problem to be solved in the belief space, so that the population space has higher evolution efficiency. The evaluate () function in the population space is an objective function (fitness function) whose role is to evaluate the fitness value of an individual in the population space. The Select () function selects a part of individuals from the newly generated individuals as parents of the next generation individuals according to the rule.

In order to make the objects, technical solutions and advantages of the present invention clearer and more obvious, the present invention is further described in detail below with reference to the accompanying drawings and technical solutions.

The embodiment of the invention provides a multi-constraint service selection method based on global QoS decomposition, which is realized by establishing a single-target optimization model with multiple constraint conditions, as shown in FIG. 1, and the realization process comprises the following steps:

101. establishing a corresponding dependency set and a conflict set for each candidate service according to the transfer characteristic of the service dependency relationship;

102. decomposing the global QoS constraint provided by the user into local QoS constraints corresponding to each service class through a CGA culture genetic algorithm; filtering candidate services which do not meet the local QoS constraint under the service class;

103. checking all the filtered candidate services, if the candidate services are in the dependence set of the filtered candidate services, removing the candidate services from the dependence set, and updating the dependence set and the conflict set of the remaining candidate services;

104. performing non-solution quality scale combination replacement by a self-adaptive replacement method aiming at candidate services in the service class, and dynamically adjusting a QoS constraint boundary;

105. calculating the service compatibility and local fitness of the candidate service based on the global QoS constraint and the functional constraint between services;

106. according to the calculation result of the local fitness of the candidate services in the step 105, the candidate service with the maximum local fitness in each service class is selected, and the final combined service is formed according to the selected candidate service.

Compared with the existing global QoS service selection method, the scheme of the embodiment is greatly optimized in terms of computational complexity and running time, and can meet the real-time requirements of users.

Service composition based on global QoS constraints and functional constraints is an optimization problem and also an NP challenge. The core of the problem is to select a candidate service for each service class, so that the formed combined service can meet multiple constraint conditions and achieve the maximum comprehensive utility value. To this end, in another embodiment of the invention, the single-objective optimization model is represented as:

wherein w is_k(k is more than or equal to 1 and less than or equal to r) is the weight of the QoS attribute k, reflects the preference degree of the user to the kth QoS attribute, and meets the requirement

CS is a composite service composed of m basic service classes Si (i is equal to or greater than 1 and equal to or less than m), and is denoted as CS { S1, S2, … …, Sm }, in this embodiment, only the sequential composite service model is considered, and other types of composite services can be converted into the sequential composite service model; x is the number of_ijRepresentative service class S_jOf the ith candidate service, x_ijWhen 1 indicates that the service is selected, x_ijWhen 0 indicates that the service is not selected,

representing a basic service class S_jAnd only one candidate service is selected; global QoS constraint relation C ═ C₁,C₂,......C_rDenotes C_kRepresenting global constraints of the kth QoS Attribute, q_k(CS)≤C_kExpressing that the k-th QoS attribute aggregation value of the combined service needs to meet the corresponding global constraint, converting the benefit type QoS attribute into the cost type QoS attribute through negative value calculation, and calculating the QoS attribute value of the combined service by utilizing a QoS aggregation formula of a sequential combined service model; the functional constraints between services are respectively defined by a dependency set T ═ T₁,t₂,......t_kD ═ D } and the set of conflicts₁,d₂,......d_uDenotes, dependency relationships

Representing service class S_aThe function of the b-th candidate service in (a) depends on the service class S_cThe d-th candidate service in (1). x is the number of_ab-x_cdLess than or equal to 0 ensures that only the candidate service s_cdWhen implemented, service s_abThis is possible. Conflicting relationships

Representing service class S_eFunction and service class S of the f-th candidate service in (1)_gThe functions of the h-th candidate service in (a) conflict with each other. x is the number of_ef+x_ghLess than or equal to 1 guarantees the candidate service s_efAnd service s_ghAnd not simultaneously in the composite service.

As shown in fig. 2, the global QoS constraint is decomposed into local QoS constraints corresponding to each service class, and candidate services that do not satisfy the multi-constraint condition are filtered in combination with the dependency-conflict relationship between the candidate services. If no feasible composite service is available after filtering, an adaptive replacement strategy is executed to dynamically adjust the QoS constraint boundary for re-filtering. And then calculating the local fitness of the remaining candidate services, and taking out the final combined service by utilizing local optimal selection. According to the transfer characteristic of the dependency relation, providing each candidate service s_iEstablishing corresponding dependent set t(s)_i) And a conflicting set d(s)_i) And providing information for subsequent candidate service filtering and local fitness calculation. In yet another embodiment of the present invention, the dependency transfer rules are as follows:

and s_d∈d(s_c) (3)

And s_b∈d(s_c)(5)

The formula (2) represents if s_bDependent on s_aThen s_bIs incorporated into s_aDependent set of (t)(s)_a) Performing the following steps; formula (II)(3) Is represented by the general formula s_cAnd s_dIf the two conflict with each other, the two conflict with each other and are included in the conflict set of the other party; formula (4) represents s_bDependent on s_a，s_aDependent on s_cThen s_bIs incorporated into s_cDependent set of (t)(s)_c) Performing the following steps; formula (5) represents s_bDependent on s_a，s_aAnd s_cConflict, then s_bAnd s_cThere is also a conflict.

Since the same QoS attribute values of different service classes are different, if all QoS constraints are directly and averagely decomposed into each service class, the selected composite service may not satisfy the global QoS constraint. To this end, in another embodiment of the present invention, the CGA cultural genetic algorithm is decomposed into local QoS constraints corresponding to each service class, a chromosome model is constructed to represent a quality scale combination, each service class corresponds to a quality scale combination, the evolution operation of the CA genetic method is embedded into the population space of the CA cultural method, and a collaborative learning mechanism is introduced into the belief space, and the global QoS constraints are decomposed into local QoS constraints as follows:

step 21, randomly generating an initial effective solution in the population space according to the size of the population space, and evaluating all solutions through a fitness function;

step 22, performing evolution operation on the solution in the population space, wherein the evolution operation at least comprises selection operation, cross operation and variation operation;

step 23, selecting a better solution from the population space, transmitting the better solution to the belief space, and replacing the worse solution in the belief space according to the cumulative update times of the solutions in the belief space; performing collaborative learning operation on the solutions in the belief space, evaluating the newly generated solutions, reselecting a better solution from the belief space, and filtering out the rest solutions;

step 24, judging whether the current iteration times meet set conditions, if so, outputting an optimal solution in a belief space; otherwise, the loop iteration is performed by returning to step 22.

Since a composite service is composed of m service classes, each service class corresponds to a quality scaleAnd combining, so that the final result of the global QoS optimal decomposition is to solve m optimal quality scale combinations. Based on this, this embodiment designs an m-dimensional chromosome coding model to represent m quality scale combinations, as shown in FIG. 5, where S is₁,S₂,...,S_mFor basic service classes, q₁,q₂,...,q_rFor the purpose of the set of QoS attributes,

representative service class S_mThe e quality scale of the r QoS attributes, each line of the chromosome represents a quality scale combination of a service class.

The Genetic Algorithm (GA) is a global search Algorithm derived by simulating evolution operations such as selection, intersection, variation and the like in nature by adopting the thought of survival, excellence and disadvantage of a suitable person in the Darwinian evolution theory, has the characteristics of simplicity, robustness, universality and the like, but has the defects of easy premature convergence and low convergence speed. The Culture Algorithm (CA) is a double-layer evolution model Algorithm proposed by Reynolds, and is composed of two parts, namely, a population space (population space) and a belief space (belief space), and simulates the evolution process of a Culture from a microscopic angle and a macroscopic angle respectively. The core idea is that evolution experience formed by a population space is selectively transferred to a belief space in an iteration process, and the belief space forms the population experience through comparison optimization and guides the evolution process of the population space, so that higher evolution efficiency is obtained. The embodiment combines the advantages and disadvantages of the two algorithms, and provides a CGA culture genetic algorithm, the basic idea is to embed the evolution operation of the genetic algorithm into the population space of the culture algorithm, and introduce a collaborative learning mechanism into the belief space, so that the optimization capability of the algorithm is improved, the problem of global QoS optimal decomposition can be effectively solved, and the framework is shown in FIG. 3. Knowing that the global QoS constraint corresponding to the kth QoS attribute is C_k，q_r(s_jn) Representing service class S_jThe nth QoS attribute value of the nth candidate service,

representing service class S_jAnd satisfies the d quality scale of the r-th QoS attribute

WhereinAnd

respectively service class S_jD is the total number of quality scales, and the decomposition process of the quality scales is shown in fig. 4.

Since the objective of global QoS optimal decomposition is to find the optimal quality scale combination for each service class, in another embodiment of the present invention, the service classes are transformed into a single-objective optimization problem without considering the difference between the merits of the service classes, and the fitness function is expressed as:

wherein

for the weight of the mass scale, the calculation formula is as follows:

for service class S_jZhongbelong to the quality staff gauge

Number of candidate services within, n (S)_j) Is service class S_jThe number of candidate services in (a) is,

is service class S_jZhongbelong to the quality staff gauge

Maximum utility function value, Q, of all candidate services within_max(S_j) Is service class S_jMaximum utility function values of all candidate services; m is the total number of service classes, r is the total number of QoS attributes, and d is the total number of quality scales; the formula (11) ensures that the decomposed local QoS constraints can also meet the global QoS constraints after aggregation; in the formula (12)

Representing service class S_jMedium mass scale

Is selected, otherwise

It is guaranteed that under each QoS attribute of each service class, only one quality scale is selected. Weight on mass scale

Is represented in the service class S_jIs selected from the kth QoS attributesThe probability of the candidate service being selected, which is calculated in the formula

For service class S_jZhongbelong to the quality staff gauge

Number of candidate services within, n (S)_j) Is a clothesClass S_jThe number of candidate services in (a) is,

is service class S_jZhongbelong to the quality staff gauge

Maximum utility function value, Q, of all candidate services within_max(S_j) Is service class S_jThe maximum utility function value of all candidate services. The optimal quality scale combination of a service class should make the number of candidate services under the combination more and better, and the utility function value more and better, so that the global optimal combined service is more and more likely to be formed, and therefore, the concept of quality scale weight is introduced.

Preferably, the cooperative learning means that the learner exchanges and shares knowledge with other people in a group through competition, cooperation, role playing and other modes, so as to effectively improve the learning efficiency. At present, most of the group intelligent algorithms cannot learn each other, so that the performance of the algorithm cannot be further improved. Based on the theory of cooperative learning, the invention introduces a learning mechanism into the belief space of the CGA algorithm, and inherits excellent genes to the next generation through the mutual learning among chromosomes, thereby accelerating the convergence rate, and in another embodiment of the invention, referring to the specific learning process of the belief space shown in FIG. 6, the cooperative learning operation is implemented on the solution in the belief space, which comprises the following contents:

step 1: randomly selecting t chromosomes from the belief space to form a cooperative learning group which is marked as

Wherein L is_i(1. ltoreq. i.ltoreq.t) represents a chromosome,

representing the jth gene of a chromosome as the service class S_jThe mass scale combination of (1);

step 2: comparative learning of genes per line of all chromosomes within a GroupAnd selecting the optimal gene of each line

I.e. from the service class S_jSelecting an optimal one of the t combinations of the quality scales;

step 3: recombining the optimal genes of each row to form a new chromosome, i.e.Each service class S of the chromosome_jThe quality scale combinations of (a) are all within Group optimal.

In step 102, a first layer of filtering is carried out on candidate services of which the service class does not meet the local QoS constraint; the second level of filtering is accomplished by examining all filtered candidate services and removing an existing candidate service from the filtered set of candidates in step 103. After the two filtering mechanisms, if there is no candidate service in a certain service class, a feasible combined service cannot be formed. The reason for this is that after the filtering of the first layer local QoS constraint condition, there may exist a condition that the remaining candidate services of a certain service class all have functional conflicts with other candidate service classes or all have dependency relationship with the filtered services, and at this time, after the filtering of the second layer functional constraint condition, these services will be all filtered, and there may occur a condition that the service class has no candidate service, and this condition is called "no solution". In another embodiment of the present invention, a solution-free quality scale combination replacement is performed by an adaptive replacement method to dynamically adjust the QoS constraint boundary, that is: after filtering, if there is no candidate service in the service class, that is, there is a situation of no solution, replacing the optimal quality scale combination of all current service classes with the suboptimal quality scale combination in the belief space, and returning to step 102 to execute until no solution-free state appears any more. After the iteration of the CGA culture genetic algorithm in the step 102 is finished, a plurality of feasible solutions exist in the belief space, so that the self-adaptive replacement method in the embodiment does not need to operate the CGA algorithm again, and only needs to directly replace the current optimal solution by using the rest suboptimal solutions in the belief space; and massive service resources exist in the cloud environment, and the probability of no solution state is extremely low, so that the execution efficiency of the whole service selection method cannot be greatly influenced by the self-adaptive replacement strategy.

After the steps 101-104, the scale of the candidate service space is reduced to a great extent. Under the conditions of global QoS constraint and inter-service functionality constraint, two factors are used for determining the degree of quality of a candidate service: its own utility function value and compatibility in the candidate service space. The larger the utility function value is, the better the compatibility is, the larger the probability of being selected as a local optimum service is. To measure this probability, in other embodiments of the invention, a candidate service s is assumed_jBelonging to the service class S_jThen, the calculation formula of the service compatibility is as follows:

wherein, t_i(s_j) Is a service s_jDependent set of (t)(s)_j) In the service class S_iThe number of services. d_i(s_j) Is a service s_jConflict set d(s)_j) In the service class S_iThe number of services. I S_iIs service class S_iThe total number of candidate services; the calculation formula of the local fitness is as follows:

u(s_j)＝con(s_j)×Q(s_j)，

，con(s_j) Is s_jService compatibility of (C), Q(s)_j) Is the value of the utility function thereof,

is service class S_jThe maximum value on the k-th QoS attribute,

is service class S_jMinimum value on k-th QoS Attribute, q_k(s_j) Is a candidate service s_jValue on the kth QoS Attribute, w_kIs the weight of the kth QoS attribute given by the user.

Service compatibility con(s)_j) Representing candidate services s_jThe overall compatibility with other candidate class services. It is served by s_jDependent set and conflicting set. Service s_jDependent set of (t)(s)_j) The larger the number of candidate services depends on s_jOnce service s is explained_jNot selected, there will be many candidate services that are not selectable. Thus, t(s)_j) Reflects the service s_jThe importance of (c). Service s_jConflict set d(s)_j) The smaller the number of candidate services and s_jConflict, statement s_jThe better the compatibility. For a Web service, the more candidate services that depend on it, the fewer candidate services that conflict with it, and the greater its service compatibility. Local fitness u(s)_j) Reflecting the candidate service s_jThe degree of quality in the service class is determined by the value of utility function and the degree of service compatibility, for s_jThe embodiment of the utility function value calculating method adopts a simple weighting function in the Alrifai M, that is, the candidate service s is calculated_jIs compared with the maximum or minimum value of the attribute in its service class, thereby normalizing all QoS attribute values to a real number interval 0,1]Within the range, and then multiplying by the corresponding weight to obtain the utility function value. Local fitness reflecting candidate service s_jThe value of the degree of superiority and inferiority in the service class to which the service belongs influences the later local optimal service selection. And selecting the candidate service with the maximum local fitness value in each service class based on the calculation result of the local fitness to form the final globally optimal or approximately globally optimal combined service.

Corresponding to the foregoing method, an embodiment of the present invention further provides a multi-constraint service selection apparatus based on global QoS decomposition, as shown in fig. 7, implemented by a single-target optimization model with multiple constraint conditions, including: a function constraint establishing module 201, a local constraint decomposition module 202, a candidate service updating module 203, a no-solution state replacement module 204, a local fitness solving module 205 and a final composite service selecting module 206,

a function constraint establishing module 201, configured to establish a corresponding dependency set and a conflict set for each candidate service according to a transfer characteristic of the service dependency relationship;

a local constraint decomposition module 202, configured to decompose the global QoS constraint provided by the user into local QoS constraints corresponding to each service class through a CGA cultural genetic algorithm; filtering candidate services which do not meet the local QoS constraint under the service class;

a candidate service updating module 203, configured to check all the filtered candidate services, remove a candidate service from the filtered dependency set if the candidate service is in the filtered dependency set of the candidate service, and update the dependency set and the conflict set of the remaining candidate services;

a no-solution state replacement module 204, configured to replace a no-solution state quality scale combination by a self-adaptive replacement method for candidate services in the service class, and dynamically adjust a QoS constraint boundary;

a local fitness solving module 205, configured to calculate service compatibility and local fitness of the candidate service according to the global QoS constraint and the inter-service functional constraint;

the final composite service selection module 206 selects the candidate service with the maximum local fitness in each service class according to the local fitness calculation result of the candidate service in the local fitness solution module, and forms the final composite service according to the selected candidate service.

In another embodiment of the present invention, the local constraint decomposition module comprises: a fitness function design unit, a chromosome model construction unit and a belief space learning unit,

the fitness function design unit is used for converting the global QoS optimal decomposition problem into a single-target optimization problem by designing a fitness function;

a chromosome model construction unit for representing quality scale combinations by constructing a chromosome model, each service class corresponding to one quality scale combination;

and the belief space learning unit is used for embedding the evolution operation of the CA genetic method into the population space of the CA culture method, introducing a collaborative learning mechanism into the belief space and acquiring the optimal quality scale combination of all the service classes.

Based on the above analysis, the CGA-based global QoS decomposition algorithm is as follows:

in order to further verify the effectiveness of the present invention, the performance of algorithm execution performance and the optimality of the combined service scheme under different candidate service scales and different constraint scales of the multi-constraint service selection method based on global QoS decomposition are verified through simulation experiments. The execution performance of the algorithm is expressed in terms of execution time in milliseconds (ms), and the quality of the composite service is expressed in terms of fitness, which ranges from (0, 1).

The experiment analyzes and compares the multi-constraint service selection method (WSD-CGA algorithm) based on global QoS decomposition with other three multi-constraint service selection methods based on QoS, wherein the first method is a global optimization service selection method (GS for short) solved by adopting mixed integer programming, the second method is a service selection method (HGA for short) under the multi-constraint relation by adopting a mixed genetic algorithm, and the third method is a multi-constraint service selection method (LDPSO for short) based on local approximate filtering. A public effective data set QWS is adopted in the experiment, and all data of the QWS comes from a public Web service on the Internet. The data set includes 2500 real Web services and their corresponding 9 QoS attribute values. The present invention assumes that the composite service consists of 5 basic service classes and selects 4 QoS attributes in the QWS dataset, respectively response time, availability, reliability andthe weight values of the cost are respectively 0.35, 0.25, 0.3 and 0.1. The global QoS constraints are: (1) response time<2s, (2) availability>0.4, (3) reliability>0.4, (4) cost<100 yuan. In the WSD-CGA method, the crossover probability is p₁0.85, the mutation probability is p₂The number q of the better solutions in the belief space is 20, the accumulated update time η of the solutions in the belief space is 5, the number t of chromosomes forming the collaborative learning group is 5, the maximum iteration time is M is 200, and the total number d of the quality scales is 10. Simulations the experiment was averaged 50 runs. The experimental environment is as follows: pentium Dual 2.4GHz, 2.0GB RAM, Windows 7, MATLAB2009a, Java 1.8.

Algorithm performance at different candidate service scales

In the experiment, the constraint scale Co among the candidate services is set as a fixed quantity, and the change of the execution time of the method and the optimality of the combination scheme is analyzed by changing the number n of the candidate services under the basic service class, wherein Co is 400, and n is changed from 40 to 360. The experimental results are shown in fig. 8: (a) to perform a temporal contrast map, (b) composite service fitness contrast. Fig. 8 shows the execution time and the fitness of the composite service of the 4 methods at different candidate service scales. As can be seen from the simulation results, the average execution time of the WSD-CGA method of the invention is 247.5ms, which is better than 592.4ms of GS, 486.3ms of HGA and 320.7ms of LDPSO, and the execution time does not increase rapidly with the increase of the number of candidate services. The WSD-CGA filters candidate service spaces through global QoS decomposition, namely the size of the search space is determined by m.r.d (m is the number of service classes, r is the number of global QoS attribute constraints, and d is the total number of quality scales), and the time performance of the WSD-CGA is better than that of other 3 algorithms because the influence of the increase of the size of the candidate service on the search space is small.

In order to compare optimality of 4 methods, the adaptability of the GS method is appointed to be f₁The HGA method has a fitness of f₂The LDPSO fitness is f₃And the fitness of the WSD-CGA is f₄Optimality is t in order₁,t₂,t₃,t₄Wherein, t₁＝f₁/f₁(optimality of Global optimization method GS)To 100%), t₂＝f₂/f₁,t₃＝f₃/f₁,t₄＝f₄/f₁。

From simulation results, the average value of the optimality of the WSD-CGA can reach 98.7%, almost approaches the optimality level of a global optimization method GS, and is superior to 92% of HGA and 97.3% of LDPSO. Because a collaborative learning mechanism is introduced into the belief space by the WSD-CGA, excellent genes are inherited through mutual learning among chromosomes, so that decomposed local QoS constraints are in line with actual conditions, and a combined service with better quality can be screened out.

Performance of algorithms under different constraint scales

In the experiment, the number n of candidate services in the basic service class is set as a fixed quantity, and the execution time of the method and the optimality change of the combination scheme are analyzed by changing the constraint scale Co among the candidate services, wherein n is 120, and Co is changed from 200 to 1000. The results of the experiment are shown in FIG. 9: (a) performing a time comparison, (b) a composite service fitness comparison. Fig. 9 shows the execution time and the fitness of the composite service of the 4 methods under different constraint scales. Because the global optimization method GS does not consider the dependency and conflict relationship among the candidate services, the execution time of the global optimization method GS is not influenced by the constraint scale, the average value reaches 286.4ms, even is better than the average value 410.6ms of the WSD-CGA method, but in a cloud computing environment, a large number of dependency or conflict relationships must exist among mass Web service resources, so that the combined service solved by the global optimization method GS has a very large probability and cannot meet the actual requirements of users. The WSD-CGA establishes a dependency set and a conflict set for each candidate service, and the process may consume more time due to the severe increase of the constraint scale, but provides more information for multi-constraint relation filtering, can filter more useless services, avoids a large amount of redundant searching, saves time for subsequent local optimization, and has time advantage compared with HGA (1740.5ms) and LDPSO (726.3 ms).

According to simulation results, when the constraint scale is in the range of 200, 1000, the average value of the optimality of the WSD-CGA reaches 97.9 percent, which is better than 93.7 percent of HGA and 96.5 percent of LDPSO. The reason is that the WSD-CGA defines the concept of compatibility when local optimal service selection is carried out, and the concept is closely related to the constraint scale between candidate services. By incorporating the multi-constraint relationship among the services into the service selection process, the finally screened combined service quality is better and can better meet the actual application condition.

The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.

The elements of the various examples and method steps described in connection with the embodiments disclosed herein may be embodied in electronic hardware, computer software, or combinations of both, and the components and steps of the examples have been described in a functional generic sense in the foregoing description for clarity of hardware and software interchangeability. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.

Those skilled in the art will appreciate that all or part of the steps of the above methods may be implemented by instructing the relevant hardware through a program, which may be stored in a computer-readable storage medium, such as: read-only memory, magnetic or optical disk, and the like. Alternatively, all or part of the steps of the foregoing embodiments may also be implemented by using one or more integrated circuits, and accordingly, each module/unit in the foregoing embodiments may be implemented in the form of hardware, and may also be implemented in the form of a software functional module. The present invention is not limited to any specific form of combination of hardware and software.

The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims (9)

1. A multi-constraint service selection method based on global QoS decomposition is characterized in that the method is realized by establishing a single-target optimization model with multiple constraint conditions, and the realization process comprises the following steps:

step1, establishing a corresponding dependency set and a conflict set for each candidate service according to the transfer characteristics of the service dependency relationship;

step2, decomposing the global QoS constraint provided by the user into local QoS constraints corresponding to each service class through a CGA culture genetic algorithm; filtering candidate services which do not meet the local QoS constraint under the service class;

step3, checking all the filtered candidate services, if the candidate services are in the dependence set of the filtered candidate services, removing the candidate services from the dependence set, and updating the dependence set and the conflict set of the remaining candidate services;

step 4, performing non-solution quality scale combination replacement on candidate services in the service class by a self-adaptive replacement method, and dynamically adjusting a QoS constraint boundary;

step 5, calculating the service compatibility and local fitness of the candidate service based on the global QoS constraint and the functional constraint among the services;

step 6, selecting the candidate service with the maximum local fitness in each service class according to the local fitness calculation result of the candidate service in the step 5, and forming a final combined service according to the selected candidate service;

in the step 5: hypothesis candidate service s_jBelonging to the service class S_jThen, the calculation formula of the service compatibility is as follows:

，

wherein, t_i(s_j) Is a service s_jDependent set of (t)(s)_j) In the service class S_iNumber of services of d_i(s_j) Is a service s_jConflict set d(s)_j) In the service class S_iNumber of services, | S_iIs service class S_iThe total number of candidate services; the calculation formula of the local fitness is as follows:

u(s_j)＝con(s_j)×Q(s_j)，

，con(s_j) Is s_jService compatibility of (C), Q(s)_j) Is the value of the utility function thereof,

is service class S_jThe maximum value on the k-th QoS attribute,

is service class S_jMinimum value on k-th QoS Attribute, q_k(s_j) Is a candidate service s_jValue on the kth QoS Attribute, w_kIs the weight of the kth QoS attribute given by the user.

2. The global QoS factorization based multi-constraint service selection method of claim 1, wherein the single-objective optimization model is expressed as:

wherein，w_k(k is more than or equal to 1 and less than or equal to r) is the weight of the QoS attribute k, and satisfies

CS is a composite service composed of m basic service classes Si (1 ≦ i ≦ m), and is denoted as CS { S1, S2, … …, Sm }; x is the number of_ijRepresentative service class S_jThe selection state of the ith candidate service; global QoS constraint relation C ═ C₁,C₂,......C_rDenotes C_kRepresenting global constraints of the kth QoS Attribute, q_k(CS)≤C_kIndicating that the k-th QoS attribute aggregate value of the composite service is required to meet the corresponding global constraint; the functional constraints between services are respectively defined by a dependency set T ═ T₁,t₂,......t_kD ═ D } and the set of conflicts₁,d₂,......d_uDenotes, dependency relationshipsRepresenting service class S_aThe function of the b-th candidate service in (a) depends on the service class S_cThe d-th candidate service in (a); conflicting relationships

Representing service class S_eFunction and service class S of the f-th candidate service in (1)_gThe functions of the h-th candidate service in (a) conflict with each other.

3. The method for selecting services with multiple constraints based on global QoS decomposition according to claim 1, wherein the step1 comprises the following steps: the dependency delivery rules are as follows:

，

formula (2) represents s_bDependent on s_aThen s_bIs incorporated into s_aDependent set of (t)(s)_a) Performing the following steps; formula (3) represents s_cAnd s_dIf the two conflict with each other, the two conflict with each other and are included in the conflict set of the other party; formula (4) represents s_bDependent on s_a，s_aDependent on s_cThen s_bIs incorporated into s_cDependent set of (t)(s)_c) Performing the following steps; formula (5) represents s_bDependent on s_a，s_aAnd s_cConflict, then s_bAnd s_cThere is also a conflict.

4. The method for selecting multi-constraint services based on global QoS decomposition according to claim 1, wherein the step2 is to decompose the CGA cultural genetic algorithm into local QoS constraints corresponding to each service class, express quality scale combinations by constructing a chromosome model, each service class corresponds to one quality scale combination, embed the evolution operation of the CA genetic method into the population space of the CA cultural method, and introduce a collaborative learning mechanism into the belief space, and the global QoS constraints are decomposed into the local QoS constraints by the following implementation processes:

step 21, randomly generating an initial effective solution in the population space according to the size of the population space, and evaluating all solutions through a fitness function;

step 22, performing evolution operation on the solution in the population space, wherein the evolution operation at least comprises selection operation, cross operation and variation operation;

step 23, selecting a better solution from the population space, transmitting the better solution to the belief space, and replacing the worse solution in the belief space according to the cumulative update times of the solutions in the belief space; performing collaborative learning operation on the solutions in the belief space, evaluating the newly generated solutions, reselecting a better solution from the belief space, and filtering out the rest solutions;

step 24, judging whether the current iteration times meet set conditions, if so, outputting an optimal solution in a belief space; otherwise, the loop iteration is performed by returning to step 22.

5. The method of claim 4, wherein the fitness function is expressed as:

，

wherein,for the weight of the mass scale, the calculation formula is as follows:

for service class S_jZhongbelong to the quality staff gauge

Number of candidate services within, n (S)_j) Is service class S_jThe number of candidate services in (a) is,

is service class S_jZhongbelong to the quality staff gauge

Maximum utility function value, Q, of all candidate services within_max(S_j) Is service class S_jMaximum utility function values of all candidate services; m is the total number of service classes, r is the total number of QoS attributes, and d is the total number of quality scales; the formula (11) ensures that the decomposed local QoS constraints can also meet the global QoS constraints after aggregation; in the formula (12)

Representing service class S_jMedium mass scale

Is selected, otherwise

It is guaranteed that under each QoS attribute of each service class, only one quality scale is selected.

6. The method for selecting services with multiple constraints based on global QoS decomposition as claimed in claim 4, wherein the step 23 of implementing a collaborative learning operation on the solution in the belief space comprises the following steps:

step 1: randomly selecting t chromosomes from the belief space to form a cooperative learning group which is marked as

Wherein L is_i(1. ltoreq. i.ltoreq.t) represents a chromosome,

representing the jth gene of a chromosome as the service class S_jThe mass scale combination of (1);

step 2: comparing and learning the genes of each row of all chromosomes in the Group, and selecting the optimal genes of each row

I.e. from the service class S_jSelecting an optimal one of the t combinations of the quality scales;

step 3: recombining the optimal genes of each row to form a new chromosome, i.e.

Each service class S of the chromosome_jThe quality scale combinations of (a) are all within Group optimal.

7. The method as claimed in claim 6, wherein the step 4 of performing non-solution quality scale combination replacement by using an adaptive replacement method to dynamically adjust the QoS constraint boundary comprises the following steps: after filtering, if the service class has no candidate service, namely the situation of no solution exists, replacing the optimal quality scale combination of all the current service classes with the suboptimal quality scale combination in the belief space, and returning to the step2 for execution until no solution state appears.

8. A multi-constraint service selection apparatus based on global QoS decomposition, characterized in that, based on the multi-constraint service selection method of claim 1 and realized by a single-target optimization model with multiple constraint conditions, it comprises: a function constraint establishing module, a local constraint decomposition module, a candidate service updating module, a non-solution state replacement module, a local fitness solving module and a final combined service selecting module,

the function constraint establishing module is used for establishing a corresponding dependency set and a corresponding conflict set for each candidate service according to the transmission characteristics of the service dependency relationship;

the local constraint decomposition module is used for decomposing the global QoS constraint provided by the user into local QoS constraints corresponding to each service class through a CGA culture genetic algorithm; filtering candidate services which do not meet the local QoS constraint under the service class;

the candidate service updating module is used for checking all the filtered candidate services, removing the candidate services from the dependency set if the candidate services are in the dependency set of the filtered candidate services, and updating the dependency set and the conflict set of the remaining candidate services;

the non-solution state replacement module is used for replacing the quality scale combination in the non-solution state by a self-adaptive replacement method aiming at the candidate service in the service class and dynamically adjusting the QoS constraint boundary;

the local fitness solving module is used for calculating the service compatibility and the local fitness of the candidate service according to the global QoS constraint and the functional constraint between the services;

and the final combined service selection module is used for selecting the candidate service with the maximum local fitness in each service class according to the local fitness calculation result of the candidate service in the local fitness solving module and forming the final combined service according to the selected candidate service.

9. The global QoS factorization based multi-constraint service selection device of claim 8, wherein said local constraint factorization module comprises: a fitness function design unit, a chromosome model construction unit and a belief space learning unit,

the fitness function design unit is used for converting the global QoS optimal decomposition problem into a single-target optimization problem by designing a fitness function;

a chromosome model construction unit for representing quality scale combinations by constructing a chromosome model, each service class corresponding to one quality scale combination;

and the belief space learning unit is used for embedding the evolution operation of the CA genetic method into the population space of the CA culture method, introducing a collaborative learning mechanism into the belief space and acquiring the optimal quality scale combination of all the service classes.

Images